Download VAST Lite Volume Annotation and Segmentation Tool User Manual

VAST Lite
Volume Annotation and Segmentation Tool
User Manual, VAST Lite RC1
Daniel R. Berger
October 20, 2014
ii
Contents
1 Introduction
1
2 Getting Started
2.1 System Requirements . . . . . . . . .
2.2 Program Setup . . . . . . . . . . . . .
2.2.1 Try It Out! . . . . . . . . . . .
2.2.2 Preferences . . . . . . . . . . .
2.3 Additional Files Included With VAST
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
5
6
6
6
8
3 Working with VAST
3.1 Image Stack Importing . . . . . . . . . . . . . . . . .
3.1.1 Importing image stacks: Pattern-based names
3.1.2 Lossless and lossy compression . . . . . . . .
3.1.3 Importing 3D volume files . . . . . . . . . . .
3.1.4 Image scale and description . . . . . . . . . .
3.1.5 The RAM usage indicator . . . . . . . . . . .
3.2 Viewing and Navigating an Image Stack . . . . . . .
3.2.1 Remote image stacks . . . . . . . . . . . . . .
3.2.2 The sidebar . . . . . . . . . . . . . . . . . . .
3.2.3 Getting and setting coordinates . . . . . . . .
3.2.4 Layers . . . . . . . . . . . . . . . . . . . . . .
3.3 Painting . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1 Multi-scale painting . . . . . . . . . . . . . .
3.3.2 Automatic Z-filling . . . . . . . . . . . . . . .
3.3.3 Using conditional painting . . . . . . . . . . .
3.4 Segments . . . . . . . . . . . . . . . . . . . . . . . .
3.4.1 Picking segments . . . . . . . . . . . . . . . .
3.4.2 The segment hierarchy . . . . . . . . . . . . .
3.4.3 Re-ordering and moving segments in the tree
3.4.4 Collapsing and expanding tree branches . . .
3.4.5 Using anchor points . . . . . . . . . . . . . .
3.4.6 Adding new segments . . . . . . . . . . . . .
3.4.7 Helper functions for arranging segments . . .
3.4.8 Select recently selected segments . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
10
12
13
13
13
14
14
15
15
16
17
19
19
20
21
21
21
21
22
22
22
23
23
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
iv
CONTENTS
3.5
3.6
3.7
3.8
3.4.9 Global operations: Deleting and welding segment
3.4.10 Segment tags . . . . . . . . . . . . . . . . . . . .
3.4.11 Editing the color of a segment . . . . . . . . . .
3.4.12 Exporting segment metadata . . . . . . . . . . .
3.4.13 Segment information . . . . . . . . . . . . . . . .
3.4.14 Searching for a segment with a given name or ID
3.4.15 The ’Collect’ tool . . . . . . . . . . . . . . . . . .
Saving Segmentations . . . . . . . . . . . . . . . . . . .
3.5.1 Save Segmentation As Special . . . . . . . . . . .
Segmentation Merging . . . . . . . . . . . . . . . . . . .
Importing Segmentations From Image Stacks . . . . . .
Exporting Image Stacks . . . . . . . . . . . . . . . . . .
A FAQ and Trouble Shooting
A.1 Frequently Asked Questions . . . . . . .
A.2 Typical Use Cases . . . . . . . . . . . .
A.3 Some Performance Tips . . . . . . . . .
A.4 Setting up VAST with a Wacom screen
A.5 Keyboard Shortcuts in VAST . . . . . .
A.6 Terms of Usage and Privacy Statement .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
subtrees
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
23
24
24
24
24
24
25
25
25
26
26
27
.
.
.
.
.
.
31
31
33
34
34
36
37
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
B Technical Information
39
B.1 Size limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
B.2 Supported file formats for importing / exporting . . . . . . . . . 39
Chapter 1
Introduction
VAST is a utility program for manual annotation and segmentation of large
volumetric (voxel) data sets. It enables users to work with voxel data sets in
the Terabyte or even Petabyte range at interactive speeds, to explore them
visually and to label structures of interest by voxel painting.
Voxel painting has a number of advantages over other manual segmentation
approaches like bread crumbing (placing a number of labeled points inside each
object) and skeletonization (placing labeled points and connecting them with
edges to form a ’skeleton’). It reveals the true shape of objects and makes
visualizations of the data more comprehensive. It also allows measurement of
volume and surface shape properties of the labeled objects. When working
on a dense segmentation, the fact that voxel painting labels areas rather than
single points or lines in the cross-sections of objects makes it a lot easier to
spot objects that have not been labeled yet (visual pop-out). It also allows
for some additional functionality, for example reliably determining neighbor
regions, using voxel overlap of several segmentations for merging regions or
determining synaptic connectivity, and conditional painting. Such functions are
also difficult to implement when using outline labeling (drawing vectorized lines
around process cross-sections in 2D slices). In addition, outline labeling can
introduce problems in identifying corresponding outlines in different slices and
can cause region overlap. Last but not least, many machine learning algorithms
for automatic segmentation rely on voxelized labelings for training, and the
output of many such algorithms are labeled voxelized regions. VAST can be
used to generate volumetric training data sets, and can to some extent also
be used for importing, proof-reading and correcting results of segmentation
algorithms.
The disadvantages of voxel painting which are mentioned most often are
that it is slower than for example skeletonization or bread crumbing, and that
it needs more storage space than alternative methods, especially when working
with very large data sets. VAST tries to alleviate both disadvantages. For voxel
painting, the time needed to label an object is largely determined by the number
of outlines that have to be drawn. VAST can do automatic convex Z-filling to
1
2
CHAPTER 1. INTRODUCTION
reduce the number of outlines by a factor of up to 8 (for standard settings) if
accurate boundary tracing is not needed. Painting is also optimized to be highly
responsive, and the user interface provides quick access to functions which are
used often (navigating through the image stack, changing the tooltip size, color
picking, and switching between paint and delete mode). To reduce the amount
of required memory and to enable interactive painting speed with any tooltip
size, VAST implements multi-scale painting.
Considering that alternative labeling methods are likely to be more errorprone and will probably require more time until an acceptably low error rate is
reached, voxel painting might be the overall faster alternative for fully manual
labeling. If an object skeleton is needed (for example to compute length of
a dendrite or number of spines etc.), it can be computed from the voxelized
segmentation, whereas the inverse operation (computing accurate volumes from
skeletons) is much harder. In a preliminary test we found that using VAST,
a very experienced user can produce a dense segmentation of well-stained and
well-aligned cortex neuropil data (6x6x30 nm voxel size) at a speed of about
4 cubic microns per hour. If cross-sections are labeled with a color dot in
the middle rather than accurately outlining and filling the cross-sections, the
labeling speed can be increased by a factor of 3 at the expense of accuracy (12
cubic microns per hour).
VAST is also very portable and light-weight. It consists of a single Windows
executable file which is independent of third-party libraries. It does not require
to be installed and can be easily copied and for example run from a memory
stick.
Many tools for labeling voxel data exist, but such tools usually have drawbacks when it comes to painting in large data sets. Most tools do not support
voxel painting but instead do bread crumbing, skeletonizing or storing object
outlines as vector graphics (for example splines). Also most of them require
the data to be loaded in RAM completely. This is not possible for the data
sets in the Terabyte (and soon Petabyte) range which are currently produced
by serial-section electron microscopy of biological samples. Many of these tools
are also developed as cross-platform applications, so that they can be run on
Windows, Mac and Linux systems. This usually means that a non-native GUI
system has to be used (e.g. Qt) which makes the program a lot more bulky and
more difficult to install and maintain. VAST is a Windows-only program and
uses native Windows GUI and graphics functions.
The key concepts of VAST are:
• Image stacks are imported into VASTs .vsv file format, where they are
stored as dices. Pre-computation of mipmaps allows for fast panning and
zooming through the data when it is opened in VAST.
• .vsv files support lossless and lossy compression to reduce the resulting
file size
3
• Image data and segmentations are stored in single files, which makes it
easy to copy them from one place to another
• Support for loading EM stacks from a web server over HTTP (openconnecto.me format)
• Dynamic multi-threaded cacheing in RAM with pre-loading for low-latency
display update
• Several image stacks can be opened and displayed together with a number
of blending and tinting options
• Multi-scale painting (in VAST, painting always happens at the currently
displayed resolution (Mip level))
• Automatic convex Z-filling during painting to speed up coarse labeling
• Automatic 2D-filling of closed contours
• Label color patterns to create a larger number of distinguishable label
colors
• Label hierarchies for fast and reversible grouping of labeled segments
• Anchor points to quickly find a given segment in the volume
• Exporting of segmentations, EM stacks and mixed image stacks (’screenshots’) in multiple formats
• Importing of segmentations from image stacks
• File-modification free editing. Image files are not changed (except if
you explicitly update information in them). Segmentation files are only
changed it you save the segmentation back to the same file.
Current limitations:
• Only one segmentation layer can be opened at a time
• Currently only 16-bit segmentations are supported
• There is no ’Undo’ function
• There is no direct export function for 3D models of segmented objects
(but this can be done by exporting the segmentation as an image stack
and using a Matlab script to generate .OBJ files from it)
• No 3D model display
• No multi-user support
• Not an open-source project, but the VAST Lite executable can be copied
and used freely (see section A.6 for details)
4
CHAPTER 1. INTRODUCTION
Please note that VAST is currently under development, and is subject to
change as new features are added and bugs are fixed. For bug reports or helpful
suggestions, please contact me at: [email protected].
Chapter 2
Getting Started
2.1
System Requirements
VAST currently only runs on 64-bit Windows computers that support DirectX
11. These are Windows Vista, Windows 7 and Windows 8 (all 64-bit), with
DirectX 11 or later installed. Windows XP and older versions will not work.
Currently only 64 bit versions of these operating systems are supported, because
32 bit programs are limited in the amount of RAM they can handle (4 GB
max. theoretically, less realistically). The computer also has to have a DirectX
11 compatible graphics card. Luckily in most modern computers even the onboard graphics chips support DirectX 11.
Recommended system configuration:
• Windows PC with 64 bit Windows Vista, Windows 7 or Windows 8
• 16 GB of RAM (the more the better)
• DirectX 11 compatible graphics card
• 2 TB of disk space (depends on the size of the data you work with)
• Wacom Cintiq 13HD or other pen touch screen with two-button pen
Minimal system configuration:
• Windows PC with 64 bit Windows Vista
• 2 GB of RAM
• DirectX 11 compatible on-board graphics card
• Standard screen and mouse
5
6
2.2
CHAPTER 2. GETTING STARTED
Program Setup
To use VAST, simply copy the executable program into a folder where you
have read/write access, and set up links on your desktop, start menu and/or
taskbar if desired. It is important that VAST has read and write access to
the folder where the executable is, because it will write a configuration file
(vast_preferences.dat) into the same folder to store your settings.
Start the executable.
2.2.1
Try It Out!
The quickest way to have a quick look at VAST is to use an online data set.
Several online data sets are included in the .ZIP package of supplementary files
as .VSVR files. You can save this package from the executable by clicking ’Yes’
in the ’First Start’ pop-up window when you start VAST for the first time, or
by choosing ’Save Documentation .ZIP To Disk...’ from the ’Info’ menu.
Unzip the package. Then, in VAST, go to ’Open EM...’ and select one of
the .VSVR files in the VAST_package/Online Datasets/ folder, for example
’openconnectome_kasthuri11.vsvr’. This will load images of a big EM stack
from the Johns Hopkins Openconnecto.me server. Your computer has to have
internet access for this to work.1
Click and drag the EM slice to pan. Use the mouse wheel to zoom. Use
UP/DOWN arrow keys to scroll through the stack.
Click on the little pencil icon in the toolbar to switch to ’paint’ mode. Choose
’Yes’ in the popup window. Click and drag over the image to paint. You can
select different paint colors in the ’Segment Colors’ tool window. To erase, hold
down the ’Delete’ key while painting (or click and hold the right and left mouse
buttons together). Select ’Keyboard Shortcuts’ from the ’Window’ menu for a
list of available keyboard functions.
2.2.2
Preferences
In the main menu of VAST, go to ’File / Preferences ...’ to open the Preferences dialog window. Here you can set the parameters for data cacheing and
display. VAST will set up the preferences for you when you run it for the first
time on a computer (whenever it cannot find the preferences file). You can edit
these preferences if you want to. You should at least check once whether the
folder in which VAST puts its temporary disk cache is on a hard drive with lots
of free space. Depending on what you do, temporary disk cache files can get
similarly large as the segmentation files you are working with, in particular if
you use global segmentation editing functions like merging, segment deleting,
or segment welding.
1 Remote EM images are loaded using the openconnecto.me cutout service over HTTP. The
VSVR file just defines the web address of the data set to load and its dimensions.
2.2. PROGRAM SETUP
7
Memory and Cacheing
On the left side you can set how much cache memory VAST will use maximally
for voxel images and for segmentations. VAST chooses initial values which are
reasonable for your system. The rule of thumb is: If you can afford it, leave 1-2
GB for the system, and split the rest 1/3 each for image cache, segmentation
cache, and general usage of VAST (don’t assign). On a system with 8 GB
RAM, this means to give 2000 MB to the image RAM cache and 2000 MB to
the segmentation RAM cache. If you are only viewing images and not using
segmentations, you can increase the size of the image cache and reduce the size
of the segmentation cache accordingly. If you plan to use other programs at the
same time or run two instances of VAST, please reduce these values as needed.
VAST will not immediately use all of the allotted memory, but it will stop
reserving new memory for cache blocks and re-use old blocks when it reaches
the limit.
In general, do not allow VAST to allocate more memory than the system
has. This can result in severe performance issues. There is a memory usage
indicator in the upper right corner of the VAST window which shows you how
much memory is currently used. The blue frame indicates the maximum amount
of RAM which VAST uses for image and segmentation cacheing. If the memory
indicator becomes red and your system slows down, try to REDUCE the cache
limits to allow Windows and VAST to use more RAM for other data.
Some of the segmentation cache is used for holding the currently displayed
part of the segmentation in memory. When you exit the preferences dialog,
VAST will tell you how much of the segmentation cache it needs for the current
display settings and whether the cache size is sufficient.
’Disk Cache Directory’: Here you can specify the folder where VAST
stores its temporary disk cache. Click the ’[ ... ]’ button to browse. Set
this to a folder where you have lots of free space (more than the size of the
largest segmentation file you will be working with, since for certain functions
VAST has to duplicate the segmentation data).
Painting
The segmentation bit depth is currently fixed at 16 bits. This allows for a
maximum of 65535 labels; however, since each segment is represented in the
tree view of the ’Segment Colors’ window in VAST, memory limitations in the
Windows system might prevent VAST from using that many labels.2
’Tablet Mode (Pen Paints, Finger Moves)’: On some pen-enabled tablet
computers VAST can distinguish finger and pen input. If this mode is enabled,
the pen will paint and the finger will move the view when in Paint Mode.
2 It
runs fine with more than 6700 labels in one of our data sets.
8
CHAPTER 2. GETTING STARTED
Display Properties
’Maximum Window Width’ specifies the width (or height, whichever is greater) of
the largest window you will be using, in pixels. This value is used to determine
how many textured tiles are needed to fill the entire window at all zoom levels.
Setting this value smaller reduces memory consumption and increases cacheing
speed, but if the value is too small, the image texture might not reach all the
way to the sides of the window at all zoom levels.
’Target Resolution Smaller Than’ lets you specify the effective resolution
of the displayed textures on the screen in screen-pixels per texture pixel. This
affects at what zoom levels which mipmaps are used. ’2’ is a good setting for
this; ’1’ makes it more detailed but slower (and more memory-consuming), and
’4’ makes it faster but blurry.
’Texture size m (texture is m^2):’ defines how large the texture tiles
will be which are used for displaying image and segmentation textures. Depending on the graphics card some texture sizes might be faster than others. I
recommend to leave this setting at 128.
’Texture Smoothing’: You can set here whether you want to use texture
interpolation. This reduces aliasing effects but can result in a slightly blurred
appearance of the textures. The most natural setting of this is, in my opinion,
’All except Mip 0’, which will show pixels with sharp boundaries only if you
zoom in more than the native resolution of the image data.
The remaining options are self-explanatory. Opacity values have to be set
between 0 (fully transparent) and 255 (fully opaque).
Press OK after you’re done configuring the preferences.
2.3
Additional Files Included With VAST
Under ’Info / Save Documentation .ZIP To Disk ...’ in the main menu
you can save out additional files which are packaged into the VAST executable,
as a ZIP file. Select a target location and save, then unzip the ZIP file.
Currently this includes a set of .vsvr files to access some large EM data
sets remotely (see section 3.2.1), some Matlab scripts which can be useful for
analyzing VAST data in Matlab, and this documentation as a PDF file.
Chapter 3
Working with VAST
VAST uses its own file format .VSV to store image data. It can open .VSV
files immediately and navigate in them quickly. If your data is a stack of for
example .PNG images, you will have to import it into VAST before you can use
it. During importing the data will be saved into a VAST-specific .VSV data file,
which allows quick access to arbitrary parts of the data. After opening a .VSV
image file, you can create a segmentation by painting on top of the images,
or you can open an associated segmentation file (.VSS) and view image and
segmentation together. VSS files tend to get big quickly, but can be packed
efficiently, for example in a ZIP file. You can view segmentations, modify and
save them. You can export segmentations as image stacks for using them in
other analysis programs or to render the segmented objects in a 3D animation
program like 3D Studio MAX. You can also import segmentation image stacks
that were generated externally.
3.1
Image Stack Importing
Typically volumetric image data is stored as a series of 2D images, or as a serial
3D block of data, which is not suitable for fast interactive viewing. When you
import such a data set into VAST, it puts the images into a single file containing
a diced data structure, and computes and includes mipmaps for the images.1
Using diced data does not only speed up loading of parts of images, but will in
the future also enable fast loading of volumetric sub-regions or 2D sections at
other orientations through the image data.
VAST does currently not include image alignment and stitching functions. If
you are starting with an unaligned stack of images, you will first have to align the
images with a different program (Fiji or Photoshop, for example) and then save
a stack of aligned images which all have the same dimensions and are named and
1 A mipmap is a downsampled version of an image. VAST uses power-of-two (2D, XY)
mipmaps. For example, for an original image of 1024x1024 pixels, it will compute mipmaps
of 512x512 and 256x256 pixels. It does this for every slice image in a stack.
9
10
CHAPTER 3. WORKING WITH VAST
numbered in a consistent way (for example img000.png, img001.png, ...).
Put all images into the same folder.
VAST can import single-tile image stacks, multi-tile image stacks, and 3D
volume files. In a single-tile image stack, each slice of the stack consists of a
single image file. In a multi-tile image stack, each slice is composed of several
tiles in a XY grid, and each tile is stored in a separate image file. A 3D volume
file stores all slice images in a single file. Currently the only 3D volume file
format that VAST supports is NIfTI (.nii). VAST will convert image data to
either 8-bit graylevel or 24-bit RGB when importing.
For importing and dicing, VAST will use the RAM cache which is normally
used for cacheing EM image data during viewing and painting. Having lots of
cache memory available will make importing somewhat faster, because images
have to be re-loaded less often. You can set the size of the EM image cache in
the Preferences (see section 2.2.2).
3.1.1
Importing image stacks: Pattern-based names
In the main menu of VAST, go to ’File / Import EM ...’. VAST will show
a file browser dialog in which you can select one or several image files. For
importing 3D NIfTI files, please select only one file. If you import a single-tile
stack and do not want to use pattern-based names, select all slice images in
the correct order, because images will be stacked in the same order in which
they appear in the system’s list of selected files. The order is usually correct if
you select the last image first, then shift-click (hold the SHIFT key down and
click left with the mouse) the first image to select the whole range. You can
also try ’Select All’ by pressing CTRL-A if the folder only contains the image
files you want to import. If you are worried about the order of the images and
want more precise control, you can use pattern-based names. If you make use of
pattern-based names to import single- or multi-tile image stacks, it is sufficient
to select one file, but even better to select the first and the last file in your set
of images. Then click ’Open’.
After selecting one or more image files (not .nii), VAST will display the
dialog shown in Figure 3.1. To import without pattern-based names, select
’Make Single-Tile Stack Using File Names and Order as Selected’ and
press OK.
If you select the second option, ’Use Pattern-Based Names’, the parameters in the lower part of the dialog window will be enabled. With pattern-based
names, you specify a template string for the file names which contains placeholders for numbers, and ranges for these numbers. With this you can also import
image stacks in which each slice is stored in several image files (multi-tile image
stacks). This is useful for data sets in which a single slice is so large that it can
not be stored in a single image file, but is stored in a set of tiles which form a
regular grid. Please note that these tiles should not be unstitched image tiles
as they come off a microscope, but they have to fit seamlessly. If you have a set
of raw microscopic images which are not yet stitched and aligned, please use an
external program to generate a stitched and aligned image stack first, and store
3.1. IMAGE STACK IMPORTING
11
Figure 3.1: First dialog for importing EM image stacks: Specification of patternbased names
each slice as a single image or a set of image tiles. You can then import those
images into VAST.
VAST will use the file(s) you selected in the previous dialog to determine
the source directory where the image files are and to generate a basic template
for the file name. It assumes that all images of the stack are in the same folder
(the one you picked an image from), and are named consistently with numbers
for slices, rows and columns. It also assumes that the set of images is complete,
which means that there’s an image for every slice/row/column combination in
the range you give. In this dialog, you specify these ranges as well as a schema
to derive the filename for a given slice/row/column coordinate.
Let’s say, for example, you have a data set called ’reallybigstack’, which
has 1000 slices, numbered from 0 to 999, and each slice has 10x8 tiles, numbered
from 1 to 10 and 1 to 8. You use a naming scheme so that the first image, in the
upper left corner of the first slice, is called ’reallybigstack_s0000_x01_y01.png’,
the image tile right of it is called ’reallybigstack_s0000_x02_y01.png’, and
so on. The last image in the lower right corner of the last slice would be called
’reallybigstack_s0999_x10_y8.png’.
First, make sure that the file name in the edit box at the top contains the
correct C++ format string (as it is used by printf()). In general, numbers
which specify the slice, column and row coordinates have to be replaced by
codes like ’%d’ (integer number) or ’%04d’ (integer number with zero-padding
to 4 digits). VAST will then fill in those numbers for each image. For more
information about format strings, refer to a C++ manual or ask the internet.
In the combo box below, select which coordinates are used in the file names,
12
CHAPTER 3. WORKING WITH VAST
Figure 3.2: Second dialog for importing EM image stacks: Image compression
options
and in which order they appear. The edit boxes below let you specify the range
of (integer) numbers for the three coordinates. After you entered all parameters,
press OK.
3.1.2
Lossless and lossy compression
Next, VAST will show a dialog where you can specify the color mode and image
compression options (Figure 3.2). Multi-file chunking and triple mip mapping
are currently not fully implemented and are therefore disabled.
Under ’Color Mode’, please select if you want to import the images as 8-bit
graylevel or 24-bit color images, and for graylevel which source color channel to
use. When importing from graylevel images, please select the first option (’from
RED channel’).
Under ’Compression’ you can specify compression options. Under ’Type’ you
can select the compression method - Uncompressed, Variable Bitdepth Compression, zlib Compression, and Spectral Compression. The three different compression algorithms are by themselves lossless, but might produce slightly smaller or
larger file size depending on your data. Variable Bitdepth Compression should
be fastest when reading from the compressed files.
’Quantization’ specifies whether the compression should be lossy or lossless.
Lossyness is achieved by quantizing, meaning throwing away bits. For example,
if you set Quantization to -2 bits, graylevel images will have only 6 bits resolution
(64 different gray levels) rather than 8 bits (256 different gray levels). Throwing
bits away reduces both file size and image quality.
’Voxel Order’ defines in which order the pixels in the images will be stored.
This can have an effect on compressed file size. ’2D Swizzle’ stores pixels in 2D
Z-order. I usually get best results using ’2D Swizzle’, but ’3D Swizzle’ might be
3.1. IMAGE STACK IMPORTING
13
superior for very well aligned data (e.g. FIB-SEM).
Next VAST will ask you to specify a target location and file name for the
resulting .vsv image volume file. Use ’.vsv’ as extension for the file name.
Choose a location where you have enough storage space for the file. The file will
not only contain the original image data, but also the mip maps. For example,
if you import 1024 images of 1024x1024 pixels each and store uncompressed, the
.vsv file will be approximately 1024·1024·1024+512·512·1024+256·256·1024 =
1409286144 Bytes (≈ 1.3 Gigabytes) large. Lossless compression will reduce the
file size, and lossy compression even more, but by how much depends strongly
on your data and the compression method used.
Then, the images will be read, diced, and put into the target file. After that,
VAST will compute the mipmaps and put those in the target file too. Depending
on the size of the data, this process can take several hours. For example, a big
data set of 350 GB takes about 5 hours to import on a recent desktop machine.
The limiting factor is the speed of hard drive access.
You can cancel the importing, but the target file will then be incomplete /
corrupted and can not be used with VAST.
3.1.3
Importing 3D volume files
Importing a 3D volume file is easier than importing an image stack. The only
format currently supported by VAST is Nifti. VAST will ask you to specify the
name of the source (.nii) file and the name of the target .vsv file. Currently
VAST requires the whole Nifti file to be loaded at once into RAM, so this only
works for smaller volumes. Also, the data in the Nifti file currently has to be 8
bit per pixel.
3.1.4
Image scale and description
After importing, you should set the voxel size of your data in the file. To
do this, go to ’Info / Volume properties ...’ in the main menu. Here
you can set the X,Y,Z size of a voxel in your data set in nanometers. Press
’Save to file’ to store the information you entered in the VSV file. This dialog
also displays how large your image stack is in voxels. The voxel size entered
here is used for the scale bar which you can enable in the main menu under
’Info / View Scale Bar’.
In the main menu under ’Info / EM File Information ...’ you can enter
and view text which will be stored in your VSV file as well. This can contain a
description of the data, copyright information, or other.
3.1.5
The RAM usage indicator
At the right side of the toolbar you can see a little field names ’RAM:’ which
shows the current RAM usage in your computer. The blue frame indicates how
much RAM VAST will use maximally for segmentation- and image cache combined. Make sure that this frame is not dedicating more than 2/3 of your total
14
CHAPTER 3. WORKING WITH VAST
RAM (you can adjust these settings in the Preferences, see section 2.2.2). The
solid blue block shows how much RAM VAST has currently allocated (including
blocks allocated for segmentation and image cache). The green area shows you
how much RAM the Windows system and other programs are using. The colors
will change to yellow if the total memory usage goes above 90%, and red if they
go above 96%. Running out of available RAM can slow down your system significantly. However, in some cases Windows uses large amounts of the available
RAM for disk caching and can free those instantly if more RAM is needed by
programs without affecting the system performance.
3.2
Viewing and Navigating an Image Stack
After you imported a stack of images, you can view them interactively. After
you closed the program, you can re-open a previously diced data set by using
’File / Open EM ...’ from the main menu. VSV files you open will be added
to a list under ’File / Open Recent EM’, from where you can quickly access
them again. The list contains the 16 most recent VSV files.
You can also open VSV (and VSS) files by drag-and-drop from a file browser
(Windows Explorer) onto the VAST window.
VAST currently has a ’Move mode’, a ’Paint mode’, a ’Collect Mode’ and
an ’Eyedropper Mode’, which you can set by clicking the tool buttons in the
toolbar. The cross of arrows icon selects ’Move mode’ and the little pencil selects
’Paint mode’. In this section we will explain how to use the ’Move mode’. For
an explanation of the other modes please refer to section 3.3.
The easiest way to navigate in the image stack is by using the mouse in ’Move
mode’. You can pan (move the image sideways) by left clicking and dragging
it. You can use the mouse wheel to zoom in and out. Alternatively, you can
zoom using the N and M buttons, or the sidebar (see below). Use the UP and
DOWN arrow keys or A and Z to scroll through the slices of the stack, or the
sidebar to scroll more quickly.
3.2.1
Remote image stacks
In addition to using an image stack in VAST which has been imported into
a local .VSV file, you can also open and access image stacks which are hosted
online. VAST supports the ’Open Connectome Project Cutout Service’ from
http://www.openconnecto.me with binary zipped data through HTTP; see:
http://www.openconnectomeproject.org/#!services/chru.2 Before you can
access the remote image stack you have to generate a .VSVR file which specifies
the parameters of the data set. .VSVR files are text files in a JSON-like format;
here is the content of the openconnectome_kasthuri11.vsvr file:
2 Essentially VAST requests [128x128x16] pixel blocks of the data set by reading
from the Open Connectome server with URLs which specify the requested region, like:
http://openconnecto.me/ocp/ca/kasthuri11/zip/6/1,129/1,129/1,17/. The received file is
then unzipped to extract the image data.
3.2. VIEWING AND NAVIGATING AN IMAGE STACK
15
{
"Comment": "Source: http://openconnecto.me/ocp/ca/kasthuri11/info/",
"ServerType": "openconnectome",
"ServerName": "openconnecto.me",
"ServerFolder": "/ocp/ca/kasthuri11",
"SourceDataSizeX": 21504,
"SourceDataSizeY": 26624,
"SourceDataSizeZ": 1850,
"TargetDataSizeX": 10747,
"TargetDataSizeY": 12895,
"TargetDataSizeZ": 1850,
"OffsetX": 0,
"OffsetY": 0,
"OffsetZ": 0,
"OffsetMip": 1,
"TargetVoxelSizeXnm": 6,
"TargetVoxelSizeYnm": 6,
"TargetVoxelSizeZnm": 30,
"TargetLayerName": "Kasthuri11@OpenConnectome"
}
VAST comes with a few pre-defined .vsvr files which you can use to open
and view some example data sets.
3.2.2
The sidebar
VAST provides a sidebar for zooming and moving through the stack. The
sidebar is a region close to the left and the right edge of the main window.
When you move the mouse cursor to the left or right edge of the window you
will see it appear as a transparent white overlay strip.3 Clicking into the sidebar
and dragging the mouse up or down will scroll through the slices of the stack
(left mouse button) or zoom (right mouse button). If you move the mouse
cursor too far away from the side of the window, the view will ’jump back’
to the previous view. If you move the mouse cursor very close to the top or
bottom of the window while scrolling (not zooming), VAST will start to scroll
continuously, with a speed depending on mouse cursor position. You can use
this function to quickly scroll through a very large image stack.
3.2.3
Getting and setting coordinates
VAST uses a coordinate system with a zero point in the upper left corner of
the first slice, with positive X to the right and positive Y down in the slice, and
Z marking the slice number. Coordinates are given in pixels at full resolution
(the coordinates are independent of the mip map displayed). The coordinates
displayed in the upper left corner of the main window show the current location
of the center of the main window. You can switch the displayed coordinates on
and off by using ’Info / View Coordinates’ from the main menu. Zooming in
or out will not move the center point of the window and therefore also not change
its coordinates. Getting or setting coordinates will also use the coordinates of
3 You
can set the opacity of the sidebar in the Preferences, under ’Side Bar Opacity’
16
CHAPTER 3. WORKING WITH VAST
the center of the screen, as do the ’anchor points’ of segments (see section 3.3).
While you drag the slice with the mouse VAST displays a transparent cross
which indicates the location of the center.4
Once you load an image stack, a tool window labeled ’Coordinates’ will
appear in the upper right corner of the main window. If the tool window is not
displayed you can open it using ’Window / Coordinates’ from the main menu.
It shows you the current center coordinates and allows you to read and set these
values. The edit field in the tool window is updated as you navigate through
the stack. To save the current location, simply copy the coordinates from that
text field (mark with the mouse and press CTRL-C), then paste it into the text
editor of your choice. You can also set the coordinates by entering or pasting
numbers here and pressing Enter. VAST will then jump to the new coordinates.
The exact format of the string does not matter; VAST simply looks for the first
three numbers in the string. VAST does not mind whether there are commas
or brackets or other non-numerical characters.
This function is quite useful if you want to store coordinates of interesting
points in an external text file or spread sheet. Please keep in mind that the
coordinate denotes the center of the current view. The center is indicated by
transparent crosshairs when you pan the view. You can also center any point by
right-clicking that point with the mouse in ’Move’ mode and selecting ’Center’
from the context menu.
The dropdown-listbox in the Coordinates tool window lists the up to 64
most recent locations you visited. A new entry is added every time you pan the
view (but currently not if you scroll through Z). You can go back to previous
locations by selecting the coordinates from this list.
3.2.4
Layers
VAST can open several image stacks at the same time, provided that they have
the same stack size. Each image stack, and also the segmentation stack, are
listed as a ’Layer’ in the ’Layers’ tool window. For image stacks (not the segmentation) the order in the list defines the order of the layers. Layers BELOW
in the list are ’in front’. The segmentation layer is always rendered on top of
all image stack layers. You can change the order of the layers by drag-and-drop
in the list. If you can not see all layers in the list, increase the size of the tool
window by dragging a corner.
Below the list of layers, the ’Layers’ tool window shows a number of ’Layer
Properties’ for the currently selected layer:
• ’Solo’: If this function is enabled, only the currently selected image layer
will be displayed.
• ’Editable’: Disabled for image layers
• ’Visible’: Transparency value for this layer. Switch off to hide layer.
4 You
can set the opacity of the center cross in the Preferences, under ’Center Cross Opacity’
3.3. PAINTING
17
• ’Bright’: Image Brightness; switch on to enable brightness control
• ’Contrast’: Image Contrast; switch on to enable contrast control
VAST can blend layers with different transparency modes. Click on the
button ’Menu’ to access more layer options. The different settings for ’Blend
Mode’ are:
• ’Flat’: All pixels in the image share the same transparency [Default]
• ’Dark Transparent’: The darker a pixel ((R+G+B)/3), the more transparent it is
• ’Bright Transparent’: The brighter a pixel ((R+G+B)/3), the more
transparent it is
• ’Max(RGB) Dark Transparent’: The darker a pixel (Max(R,G,B)), the
more transparent it is
• ’Max(RGB) Bright Transparent’: The brighter a pixel (Max(R,G,B)),
the more transparent it is
’Color Filter ...’ will open a color selection dialog where you can choose
a color by which the layer images should be filtered during display. To not filter
the images, choose white (255,255,255) [Default].
3.3
Painting
The main function currently provided by VAST is painting of segmentations as
a colored overlay of the image data. When a stack of EM images is loaded, you
can enter ’Paint Mode’ by clicking the little pencil icon in the toolbar. When
you start a new segmentation like this, VAST will ask you if you want to add
16 segments (label colors) to your segment list. Also, two floating tool windows
will appear at the right side. The upper one, ’Drawing Properties’ (Figure
3.3), provides options for drawing, whereas the lower, ’Segment Colors’, lets
you select and organize the segment labels and their colors in the segmentation.
When in paint mode, you can paint on top of the currently displayed EM
image. Select a color (label number) from the ’Segment Colors’ window at the
right by clicking on it. Then click the left mouse button where you want to
paint in the image.5 You will see the outline of your current tooltip as a circle.
By clicking and dragging the mouse you can paint larger regions. All painting
happens in an overlay plane which is blended over the EM image (the EM image
itself will not be changed). You can use the ’Alpha:’ checkbox in the toolbar to
switch the painted overlay on and off, and the slider right of it to set the opacity
of the painted overlay. Most colors are not solid colors, but have patterns. Use
the ’Pattern:’ checkbox to switch patterns on and off, and use the slider right
5 Even
though you can use VAST with a mouse, it is designed to be used with a pen tablet.
18
CHAPTER 3. WORKING WITH VAST
Figure 3.3: The Drawing Properties tool window
of it to manipulate the contrast of the patterns. If you enable the ’SelAlpha:’
checkbox, the opacity of the selected segment and its children will be controlled
separately by the SelAlpha slider. You can use this to highlight a particular
segment or set of segments. You can also switch the EM image layer on and off,
by clicking on the ’EM’ checkbox in the toolbar. This is sometimes useful if you
want to inspect just the segmentation.
You can change the size of the pen tooltip. The easiest way is, if you are
using a pen tablet and VAST is properly configured, to hold down one of the
pen buttons and to move the pen up or down on the screen. You can also use
the - and + buttons. The current pen diameter is displayed in the Drawing
Properties tool window. The third way of changing the tooltip size is to edit
the ’Pen Diam.’ text field in the tool window. You can also lock the current
tooltip size if you don’t want it to be changed accidentally, for example if the
size of the tooltip is important for your data analysis, by switching on ’Lock’.
The checkbox ’Fill’ next to it switches automatic filling of closed contours
on and off. If enabled, after each paint stroke VAST checks a rectangular area
with the approximate extent of the stroke for empty closed contours of paint
color and fills them with the paint color.
Below you can choose from ’Paint All’, ’Background’ and ’Parent’. This
determines which voxels in the current segmentation are paintable. If you select
’Paint All’, you will paint over or delete anything, no matter if it was painted
before or not. If it is set to ’Background’, previous paints will not change, but
your paint will only be applied to voxels which have not yet been painted to. If
you erase, only the current paint color will be erased to empty (background).
This is the most useful painting mode.
Instead of only affecting background pixels, ’Parent’ mode will affect only
3.3. PAINTING
19
pixels which have the color of the immediate parent of the current paint color
(see below for a description of segment hierarchies). When erasing, voxels with
the current paint color will be changed back to the parent color. This mode is
only useful in special cases, in particular when re-labeling a previously painted
area to a new color.
3.3.1
Multi-scale painting
A specialty of VAST is that it allows you to paint at different resolutions. In
fact, VAST limits you to always paint at the currently displayed resolution.
The advantage of this is that the amount of data that has to be manipulated
when you paint a stroke is limited by the window size and screen resolution.
Otherwise, for very large volumes one could easily get into a situation in which
the amount of data that has to be written for a paint stroke is much larger
than what can be loaded in RAM at one time, which would cause all sorts of
problems, including very slow painting. Also it does not make sense to paint at
a resolution which is much higher than the screen resolution because mouse (or
pen) precision is also limited. Finally, allowing low-resolution painting can save
a lot of memory, if large objects are painted coarsely.
In VAST, images are stored as a pyramid of mipmaps with reduced resolution using powers-of-two factors. Painting always happens at the resolution
of the currently displayed mipmap. This means that you can change the resolution at which you are painting by zooming. A single segmentation can be
composed of parts at different resolutions. For example it is possible to draw
a rough outline of an object at a low resolution, and then to zoom in and correct the object’s shape at a high resolution. VAST will automatically upscale
and downscale the displayed segmentation as you zoom, but zooming will not
change the painted segmentation. The segmentation is stored at the resolution
at which it was painted. If you paint at a low resolution first and then correct
at a high resolution, part of the low-resolution segmentation will be replaced by
a high-resolution version. If you paint at a high resolution first and then correct
at a low resolution, part of the high-resolution segmentation will be replaced by
a low-resolution version, including pixels in the vicinity.6
Sometimes you might want to make sure that a painted segmentation has a
certain resolution. You can enforce painting at only one resolution by restricting
painting to a particular mipmap (’Restrict’ in the Drawing Properties tool
window). VAST will then enable painting only when the image stack is zoomed
to display the selected mipmap.
3.3.2
Automatic Z-filling
The time that has to be spent to manually paint a segment in VAST depends
largely on the number of 2D outlines that have to be drawn. Especially if you
6 I have to do this because at the time of painting at a low resolution, not all higherresolution images may be available in RAM (they may even be too large to be loaded in
RAM).
20
CHAPTER 3. WORKING WITH VAST
follow a process that runs vertically through the volume, you have to paint
(almost) the same outline over and over again, for every slice. If you want to
just get a rough outline of an object and you’re not interested in a high precision
of the boundary, you could increase the painting speed by a factor of n if you
paint the outline only in every n-th slice, or paint n slices at a time. In the first
case, you get gaps of n − 1 slices between the painted outlines, in the second
case it is hard to determine what you are actually painting because you can’t
see where your color goes in most of the slices.
VAST uses a third method. It supports automatic Z-filling of intermediate
slices where the regions of the lower and the upper painted region overlap. It
turns out that in most cases neuronal objects are locally convex. Exceptions
are branches, for example when a spine neck runs very close to the dendritic
shaft. Automatic z-filling will only fill in the volumes of the overlap between
the specified painted regions, and in most cases (for convex objects) the filled
regions will stay inside the segmented object.
Z-filling makes sense across a few slices only, because there will be no overlap
if your object moves too much from slice to slice (runs oblique). Also, VAST
has to load multiple slices in RAM to be able to fill in those slices. The maximal
distance across which VAST lets you fill in depends on the size of the image
cubes used. Currently the cubes are set to be 163 voxels large, and VAST allows
you to fill in up to +-8 slices (because it loads two layers of cubes at a time). In
the data sets we are using this is approximately as far as z-filling makes sense,
and it speeds up painting by a factor of 8.
You can set how far the z-filling will reach by setting ’Max Paint Depth’ in
the Drawing Properties tool window. This value controls both the distance
at which Z-filling occurs, and the stepping distance for navigating with S, X or
PageUp, PageDown keys, to ensure gap-free painting.
Automatic Z-filling is only applied while painting, not when erasing. This
makes it easier to correct what has been filled in in the case of non-convex
neighborhoods. This also means that the best strategy to draw an object coarseto-fine is to try to paint conservatively (try to stay within the object boundaries),
and correct by adding paint rather than removing paint, because deleting has
to be done in every slice individually.
’Z-Scrolling During Paint Stroke’ is by default disabled to prevent painting errors when accidentally switching to the next slice before the paint stroke
is finished. However, if you enable it, you can very quickly coarsely label a long
neurite running through your stack vertically by scrolling through the stack
while following the neurite with your pen – provided that loading of the image
stack keeps up with the update rate of the screen.
3.3.3
Using conditional painting
The last section of the Drawing Properties tool window handles the settings
for ’conditional painting’. If you switch on conditional painting by clicking
the check box Enable, only pixels will be painted for which the EM image
fulfills certain criteria. You can choose from three methods which determine
3.4. SEGMENTS
21
the paintable pixels depending on whether the (normalized) brightness of the
image pixel (of the selected image layer) is in a certain range, which you can
set. The value range for minimum and maximum brightness is 0..1. If you
are using several image layers and conditional painting does not seem to work,
please make sure that the correct image layer is selected in the ’Layers’ window.
Currently conditional painting does not work very well with z-filling. I recommend not using z-filling (set it to 0) when you use conditional painting.
3.4
3.4.1
Segments
Picking segments
You can select the segment color to paint with in the ’Segment Colors’ tool
window by clicking on it in the tree view. You can also pick any color you
see in the segmentation layer by using the pipette tool. To do this hold down
the SHIFT key and click on the segment you wish to select. This makes is very
easy to switch between segment colors while painting. Alternatively you can use
the Pipette mode which you can select in the main toolbar. If you hover over
segment colors in the main window when in picking mode, VAST will display
the name of the segment as a tooltip.
3.4.2
The segment hierarchy
VAST can arrange segments in a tree-like hierarchy. This means that each
segment can have other segments as children, which can themselves have children, and so on. VAST also allows you to collapse and expand parts of the
tree dynamically, so that you can quickly switch between a visualization which
shows a whole branch of the tree in the same color or individual sub-branches
in individual colors. For example, if all spines of a spiny dendrite are labeled
as sub-objects (children) of the dendritic shaft, one can instantly flip between
a display in which the whole dendrite has the same color, or each spine has a
different color, by opening and closing the dendritic shaft folder. Segments can
also be used as folders to group segments, for example to classify labeled objects.
You can use tags to designate certain segments as folders, to help external analysis (see section 3.4.10). The grouping can also be applied when segmentations
are exported.
Segment hierarchies are visualized and edited in the ’Segment Colors’ tool
window. This tool window uses a ’tree view control’, similar to the navigation
pane of a windows explorer window, which makes usage very intuitive. Most
advanced functions can be found in the tool window’s menu, which opens either
by clicking the ’Menu’ button or right-clicking into the tool window.
3.4.3
Re-ordering and moving segments in the tree
To re-order the segments, simply drag and drop them with the mouse. You can
only select and drag one segment at a time, but if the segment has children
22
CHAPTER 3. WORKING WITH VAST
the whole branch will be moved (including all children). Please note that to
make a segment the first child of another segment, you have to drag it to the
right side of the tool window, right of the new parent segment. The new parent
segment will then be highlighted in blue, instead of the black line indicating the
target space between two segments. You can move any segment, with exception
of the ’Background’ segment. The Background segment can also not have any
children.
Because it can be cumbersome to move hundreds of items from one folder to
another one by one, VAST currently supports two functions ’Make all siblings
children’ and ’Make all children siblings’ which can help in certain situations
(see section 3.4.7).
3.4.4
Collapsing and expanding tree branches
You can collapse and expand tree branches, which are displayed in the same
way as folders and subfolders are in the Windows explorer, by clicking on the
little ’+’ or ’-’ sign left of parent segments. When you collapse a folder, in
the segmentation layer all its children will be displayed in the same color as the
parent. If you pick a segment color from the segmentation layer by shift-clicking,
and the selected segment is in a collapsed folder, the folder will be automatically
expanded to show the native color of the segment you selected.
3.4.5
Using anchor points
Each segment has an ’Anchor Point’ stored with it. This is an XYZ coordinate vector which indicates the location of the segment in the stack. Initially the anchor point is set to the point at which the segment is painted first.
You can jump to the anchor point by right-clicking on a used segment in the
’Segment Colors’ tool window and selecting ’Go To Anchor Point’ from the
context menu. You can quickly jump to the anchor point of the selected segment
by pressing the ’Home’ key.7 You can also set the anchor point of the selected
segment to the current view location (as indicated by the center cross) by selecting ’Set Anchor Point’ from the context menu. You will have to confirm
this action in a pop-up window to prevent accidental setting of anchor points.
3.4.6
Adding new segments
There are several functions to add segments in the context menu of the ’Segment
Colors’ tool window. You can add a segment as next sibling or as a child of the
selected segment. ’Add 10 Segments’ will add 10 segments immediately after
the selected segment. ’Add Skeleton Segments’ adds a set of child segments
to the selected segment which can be used for rudimentary skeletonization. I
have not found this function particularly useful though.
7 If you pressed the ’Home’ key accidentally and want to go back to where you were, you
can select the previous location from the drop-down menu in the ’Coordinates’ tool window.
3.4. SEGMENTS
23
The most sophisticated way to add segments is ’Add Named Segments ...’,
which lets you specify a naming scheme and add multiple named segments at
the desired target location in the segment tree. VAST will attempt to guess a
naming scheme from the name of the currently selected segment.
You can change the name of any segment in the same way as file names
are changed in the Windows Explorer – click a selected segment name a second
time, then rename it.
3.4.7
Helper functions for arranging segments
Under ’Arrange’ in the context menu you can find two useful functions to move
many segments at once. ’Make All Siblings Children’ will move all siblings
of the selected segment into its folder (make them children of the selected segment). ’Make All Children Siblings’ moves all children of the selected segment out of its folder and makes them siblings. Be careful with these functions
because currently there is no ’Undo’.
3.4.8
Select recently selected segments
Under ’Select Recently Selected’ in the ’Segment Colors’ tool window’s
context menu you can find a list of the segments you had recently selected. You
can click on one of the listed segments to select it again.
3.4.9
Global operations: Deleting and welding segment
subtrees
Deleting and welding segments is actually more difficult than it seems because it
involves traversing the whole segmentation data set and inspecting every single
painted voxel. When you choose to delete the selected segment and its children,
VAST will actually have to not only set all voxels with those segment numbers
to 0, but also renumber all the other voxels so that the used label numbers will
have no ’gaps’ (all label numbers between 0 and n are used).
’Welding’ will make the selected segment and all its children the same label.
It will renumber all voxels with label numbers of children of the selected label
to the same number as the selected label, and also, similar to when deleting,
renumber the other segment voxels so that the resulting segmentation is free of
label number gaps.
Because these functions change almost the complete segmentation and VAST
is designed so that the opened segmentation file is not changed, it basically has
to copy almost the complete segmentation data to the temporary cache file.
Depending on what segmentation you are working on this can take a lot of
time, RAM and disk space. Also it will change the internal ID numbers of
the segments, so please think twice before using these functions in case you are
relying on absolute segment IDs in your analysis. These functions are also not
well tested; please report any bugs you might encounter.
24
3.4.10
CHAPTER 3. WORKING WITH VAST
Segment tags
Each segment can have a ’Tag’ which you can select in the ’Segment Colors’
context menu under ’Tags ...’. A tag is a number between 0 and 15 which can
indicate the type of the segment. The tag value is exported with the Segment
Colors text file and can be used to help external analysis. By default the tag of
all segments is 0. VAST uses tag 1 to indicate that the segment is a ’Folder Segment’, which is not a labeled object by itself but rather a folder which contains
other folders and objects. This information can be used to collapse all objects,
but expand all folders – select ’Expand Only Segments Tagged as Folder’ under ’Expand / Collapse Child Folders’ in the ’Segment Colors’ tool window context menu.
Segments which have a tag that is not 0 will have an icon with a colored
frame in the Segment Colors treeview. Folder segments have a gray frame.
3.4.11
Editing the color of a segment
The color and pattern of any segment can be changed. Right-click on a segment
and go to the sub-menu ’Colors’ of the context menu, then choose the desired
option. Each segment has a primary color, a secondary color, and a pattern
that is used to blend between them. You can also randomize the colors of all
segments or set the primary or secondary color of all segments to the same color
(not recommended). Please remember that by collapsing segment folders you
can quickly and reversibly switch the displayed color of a segment to the color
of the collapsed parent.
3.4.12
Exporting segment metadata
The entry ’Save Segment Colors...’ in the context menu lets you export the
metadata of the segments to a text file, which you can then for example parse
with MATLAB to extract colors, hierarchies, names, anchor points etc. of the
segments for analysis.
3.4.13
Segment information
’Segment Info’ in the context menu opens a window which shows you the internal information associated with the selected segment. You can use this information to count children of the segment, get its internal ID or other parameters.
The text can be copy-pasted if needed.
3.4.14
Searching for a segment with a given name or ID
At the top of the ’Segment Colors’ tool window is an edit field which can
be used to find segments. As you type or paste a text string into this field,
VAST will select the next segment (after the selected segment in depth-first
search order) the name of which contains the typed sub-string. The edit field
3.5. SAVING SEGMENTATIONS
25
is case-sensitive. If there are more than one segment which contain this substring, you can click on the magnifying glass right of the text edit field to
go to the next segment that matches your text. The F3 key has the same
function, provided that the segment tree sub-window of the ’Segment Colors’
tool window is active.
You can also search for a segment with a particular ID. To do this, type an
opening bracket [ into the find edit field, followed by the ID number you are
looking for.
3.4.15
The ’Collect’ tool
The ’Collect’ tool (’Collect Segment Mode’) can be selected in the toolbar by
clicking the icon with an arrow pointing at a folder. When in Collect mode,
segments you click will be moved in the segment tree to become children of
the currently selected segment. This can be useful to quickly classify different
objects into different types. Simply make a folder segment for each type, select
it, and click on the objects in the image that are of that type with the ’Collect’
tool.
Using this tool is a bit dangerous because there is no Undo. If you click the
wrong object it might be difficult to remember where it came from and there is
currently no easy way to ’move it back’. Secondly, when an segment is moved,
all its child segments are moved with it, but not its parent(s). So if you are
dealing with objects which consist of several parts, make sure that you move
the parent segment of the object and not only a side branch of its tree.
3.5
Saving Segmentations
Important: VAST DOES NOT SAVE AUTOMATICALLY while you paint.
Your tracings will be held in RAM and/or a cache file on disk until you explicitly
save them. If you open a segmentation from a .VSS file and work on it, the
file will not be changed unless you explicitly tell VAST to save the changes
you made to the opened file by selecting ’Save Segmentation’ from the main
menu. If you want to keep the previous version and save to a new file, use
’Save Segmentation As ...’ instead. VAST will then take all data from the
opened file, the RAM cache, and the segmentation cache file, and combine them
into a new file on disk.
We have had cases in which people had VAST open for several days without
saving and lost a lot of work when the computer crashed. Please save your work
once in a while.
3.5.1
Save Segmentation As Special
’Save Segmentation As Special ...’ provides you with two functions to save
your current segmentation to a new file in a modified way. First, you can choose
to save only the selected segment or subtree of the current segmentation to a
26
CHAPTER 3. WORKING WITH VAST
new file. Alternatively, you can change the resolution of your segmentation
and adjust the canvas size on which the saved segmentation will open (this is
currently only supported when saving all segments).
To save only the selected segment or the selected segment and its child tree to
a new segmentation, select that option from the drop-down list and press ’Save’.
Please be aware that the internal IDs of the segments in the new segmentation
file will be changed to maintain a gap-free list of IDs from 0 to n for n segments.
The settings in the lower part of the dialog provide limited functionality to
adjust the resolution of the saved segmentation and target canvas size.
Normally VSS files only open on top of an image stack of the same size in
voxels. You can use this function to save an existing segmentation so that you
can open it on a stack which has slightly different size and/or is scaled up or
down by a power of two.
Currently the saved segmentation will stay aligned to the upper, left, top
corner of the stack. Also, only powers-of-two scaling is possible. If you want to
translate the segmentation to a different location or scale in the target stack,
you’ll have to export the segmentation as an image stack, modify the images
accordingly, and re-import.
3.6
Segmentation Merging
If you have two or more (for example partial) segmentations (.VSS files) of
the same image stack you can merge them into a single segmentation. For
this, select ’File / Merge Segmentations in ...’ from the main menu, and
choose the .VSS file(s) you would like to merge in with the currently opened
segmentation. During merging, VAST will add the selected .VSS files to the
current segmentation as if someone painted them on top in ’Paint All’ mode.
You can select several .VSS files at once for merging, which will then be added
one-by-one. Segment IDs of the files which are merged in will be changed so
that they do not overlap with existing segments.
VAST does not save the merged segmentation automatically nor does it
change any of the segmentation files. It will generate the merged segmentation in
the segmentation cache, and if it does not fit in RAM it will end up in the temporary segmentation file. You will have to save the resulting merged segmentation
if you want to keep it, for example via ’File / Save Segmentation As ...’
from the main menu.
3.7
Importing Segmentations From Image Stacks
VAST has some – not very well tested – functionality to import segmentations
from image stacks. Just like the image files generated during segmentation
exporting (see section 3.8 below), the RGB values of the image pixels should
encode the segment numbers (the least significant 8 bits (0..7) are in the blue
3.8. EXPORTING IMAGE STACKS
27
channel, bits 8..15 in the green channel, and bits 16..23 in the red channel).
Please remember that VAST can currently only handle 16-bit segmentations.
To import, select ’File / Import Segmentation ...’ from the main menu.
VAST will ask you to select one or several image files of the image stack you
want to import. Then it will open a dialog where you can specify the parameters
for segmentation importing.
File name(s):
You have to specify a name template for all image files in the stack (see section
for more information on the format of the string). You can also specify the order
and the value limits for the parameters of your file name template, to define the
names of all the images or image tiles in your stack.
Image Parameters:
Here you can rotate and flip the images if necessary, and tell VAST where to
put them into the currently opened volume. The ’Tile Size’ is extracted from
the image file you selected, but you can adjust it here too.
Segment Label Parameters:
If you have a segment metadata file for the imported segmentation stack (same
format as the file written in section 3.4.12) you can provide it here. The options
below specify how VAST should deal with the segment IDs in the imported
stack.8
Similar to segmentation merging, VAST will not save the segmentation after importing automatically. You can do that yourself after importing, using
File / Save Segmentation As ...’ from the main menu.
3.8
Exporting Image Stacks
Figure 3.4 shows how we currently analyze and visualize manual segmentations
done in VAST. Segmentations can be exported as image stacks, and these image
stacks can then be imported into MATLAB, where we compute properties of
the segments and generate mesh surfaces. MATLAB can export meshed surfaces to the common .obj file format, which we load into 3D Studio MAX for
visualization.
Go to ’File / Export ...’ in the main menu to export (parts of) the image
stack and/or segmentation stack as a stack of image files. The dialog shown in
Figure 3.5 will pop up. You can export EM image stacks, segmentation image
stacks, and screenshot image stacks. You can specify a region of the data set to
be exported, and a resolution (mip level) for the images. You can also export a
data region which is too large for storing the whole slice in a single image as a
tiled set of images.
8 Not
all options might work. Some restrictions apply.
28
CHAPTER 3. WORKING WITH VAST
Figure 3.4: Current pipeline for exporting from VAST
Export as:
Choose here if you want to export an image stack as a single-tile stack (one tile
per slice) or a multi-tile stack (a grid of image tiles per slice). For the second
option you can define the tile size to be used.
Region to export (Specify coordinates at full resolution):
This defines the region of the stack that should be exported, for all three targets
(Screenshot, EM (image) data, Segmentation). By default it is set to the full
stack. You can restrict the export region here. X is the horizontal axis in the
slice, pointing rightwards; Y is the vertical axis in the slice, pointing downwards
on the screen, and Slice (Z) defines the range of slices that should be exported.
The first and second columns of the edit fields let you define start and end of
the region for each axis, the third column defines the size of the exported region
(edit fields change each other to stay consistent).
If you click the ’Full’ button, the values will be set back to the full extent
of your stack.
The ’BBox’ button sets the region to the bounding box of the currently
selected segment. You can use this function to define a cut-out region from a
painted segment;9 you can also use a new segment and just paint the upper
left top and lower right bottom corners of the cutout region, select it and go
to ’Export Data’. Then click the ’BBox’ button to use the painted extent as an
export region.
Under ’Resolution:’ you can select at which mip level you want to export.
9 Note that the bounding box is not always correct, in particular if you delete parts of what
was painted before, the bounding box will not shrink.
3.8. EXPORTING IMAGE STACKS
29
Figure 3.5: Export dialog
VAST does not support arbitrary scaling, but can export image stacks at its
native mipmap scales (which are powers of two). You can also subsample the
stack by slices (every nth slice).
Below you can see the image size resulting from your settings and an estimate
of the (raw) data size that will be exported. Compressed image formats like
.PNG can however produce much smaller file sizes, depending on the image
content.
Screenshot target:
If you want to export a stack of ’screenshots’ how the images look in the main
window of VAST, enable the checkbox ’Export Screenshot as’. VAST will
reproduce the pattern, blending and tinting settings as they are currently set
in the main window in the exported ’screenshots’. Select a target image format
and filename prefix / location.
EM data target:
This saves a stack of (EM) image data from the selected layer. You can specify
the target format, filename prefix and location.
Segmentation target:
This saves the segment IDs of the segmentation layer or part of it as an image
stack. The segment ID of each pixel (a 16-bit number) will be encoded in the
color of the pixel in the exported image. Bits 0-7 will end up in the blue channel
and bits 8-15 will end up in the green channel. The red channel will currently
stay 0.
When you are done setting up the parameters for the export, press ’Export’
and VAST will start exporting the image stack or stacks (VAST can export
more than one target at the same time).
30
CHAPTER 3. WORKING WITH VAST
Appendix A
FAQ and Trouble Shooting
A.1
Frequently Asked Questions
My image stack is not aligned. How do I get it aligned into VAST?
VAST does not have any stack alignment (not stitching) functionality. You’ll
have to use other programs to render an aligned image stack (for example Adobe
Photoshop or plugins in Fiji [2], [4]), and then import that aligned stack into
VAST; or you’ll have to work on an unaligned image stack.
Can I analyze multi-channel optical image stacks in VAST?
Yes. You can load several image stacks at once, provided they are the aligned
and the same size, and each one can be RGB or graylevel. In addition you can
tint different image stacks in different colors to distinguish different channels.
Does VAST support 4-dimensional data sets (for example time-lapse
data of a 3D structure)?
No.
How do I open a .VSS file without a matching .VSV file in VAST?
You can not. Actually, you can, with a work-around. If you load any .VSV and
then open the .VSS in question, and the .VSS file has a different size, VAST
will tell you its dimensions (size in pixels) in the error message window. Note
down these numbers. Now create a dummy .VSVR with these dimensions (VSVR
files are just text (.txt) files with a different file name extension). Set the field
’ServerName’ to an empty string to make it a dummy layer. Then restart VAST,
open the dummy .VSVR as image stack, and open your .VSS (which should have
the same size) on top. Here is an ’empty dummy’ .VSVR example:
{
"Comment": "Empty Dummy Layer",
"ServerType": "openconnectome",
"ServerName": "",
31
32
APPENDIX A. FAQ AND TROUBLE SHOOTING
"ServerFolder": "",
"SourceDataSizeX": 49152,
"SourceDataSizeY": 32768,
"SourceDataSizeZ": 255,
"TargetDataSizeX": 49152,
"TargetDataSizeY": 32768,
"TargetDataSizeZ": 255,
"OffsetX": 0,
"OffsetY": 0,
"OffsetZ": 0,
"OffsetMip": 0,
"TargetVoxelSizeXnm": 6,
"TargetVoxelSizeYnm": 6,
"TargetVoxelSizeZnm": 30,
"TargetLayerName": "Empty Dummy Layer"
}
How do I deal with self-touching objects?
If you need to be able to recover the true shape of an object, for example for
correct skeletonization or computation of the surface area, places where there
are self-touches (for example, a dendritic spine touching the dendritic shaft) can
be problematic. One way to get around this is by using sub-objects and glue.
Just like a plastic model which is constructed from parts, you would make the
spine a child of the parent and add ’glue’ – a different segment which you treat
specially in the analysis – to the interface where parent and child object are
actually connected.
How do I make shiny 3D pictures and animations from the segmentations I painted in VAST?
VAST does currently not have 3D rendering capabilities. We use Matlab scripts
to extract .OBJ (Wavefront OBJ) model files from exported segmentation image
stacks (see section 3.8). These models can then be loaded into 3D rendering
programs (we use Autodesk 3dsMax).
Please contact me at [email protected] if you are interested in
these scripts.
Suddenly all internal segment IDs changed – What happened?
VAST currently keeps a continuous list of segment IDs between 1 and n for
n segments at all times. This means that if segments are removed from a
segmentation, the other segments will ’move up’ to keep the list continuous.
This happens when you use the functions ’Delete Segment + Subtree’,
’Weld Segment Subtree’ or save a new .vss file with a subset of the current
segments using the ’Save Segmentation As Special ... function. If you are
using the internal IDs to identify particular segments and do not want them to
change, avoid those functions. For example you can put deleted segments into a
’Deleted’ folder and/or re-use them. If you use ’Merge Segmentations in...’
or with certain settings ’Import Segmentation ...’, their internal segment IDs
will also be changed so that they don’t overlap with existing IDs.
A.2. TYPICAL USE CASES
33
Why is it called ’VAST Lite’ and not just ’VAST’ ?
The ’Lite’ in the name emphasizes that this is not supposed to be the final
version of the software. It is currently a usable tool with a limited set of capabilities, which is provided to the scientific community without restrictions.
Development of VAST continues, and there may at some point be a released
version with more features.
A.2
Typical Use Cases
Analyzing synaptic connectivity in a cortex EM stack
It is possible to define the location of synapses and their synaptic partners by
painting the synaptic membrane. We do this in two steps. First, we paint the
axons and dendrites with individual colors. Then we save this data set, and
generate a second data set in which just the synapse membranes are labeled, or
in which the synapse membranes are over-painted with a special synapse color,
with a pen with fixed size (for example 16 pixels diameter). Make sure that the
3D region painted for each synapse is a single connected component and that
different synapses are separate connected components. We then export both
data sets as image stacks and use a Matlab script to find each synapse by doing
connected-component analysis. For each synapse we then find the axon and
dendrite which occupy the same voxels as the synapse in the other data set,
which gives us the connectivity information.
Counting and classifying objects by painting
Just as for the synapses in the previous example, you can use connectedcomponent analysis in Matlab to count other objects in the stack, for example
neuron cell bodies. You can use paint all neurons of one type in the same color
and use different colors for different types. Connected-component analysis can
be used to separate the different cell bodies for each type, given that they are
separated in space. The connected-component analysis will also give you the
number of objects of each type, and their volume, if you count the painted voxels
for each connected component.
Tracking objects in a video
If you translate a video into a sequence of images, you can of course import
this image sequence as a stack into VAST (even in color). In the same way
as you can label three-dimensional structures in VAST, you can label objects
or regions or fiducial points as they move through the video. You can then
export the labelings as an image stack and analyze locations in the image and
movement.
Defining fiducial points in an unaligned image stack for manually
aided alignment
Some EM image stacks are difficult to align with automatic methods, for example if the image quality is bad, there is high-contrast background, or the tissue
34
APPENDIX A. FAQ AND TROUBLE SHOOTING
slices have folds. Manually defined fiducial points which should end up in the
same place from slice to slice can help improve the alignment. You can load an
unaligned stack into VAST and use manual painting to define fiducial points.
Use a different color for each feature you are tracking through the stack, so that
in the analysis you know which points belong together.
A.3
Some Performance Tips
• If file access is very slow (when you move through the stack and it takes
time until the images appear) consider storing the data locally and/or on
SSD drives. In particular, especially when using non-SSD drives, put files
which you use together on physically separate drives. The problem is often
that two files (for example an EM layer and the segmentation layer) are
loaded at the same time from the same non-SSD hard drive, which causes
the read/write head to jump forth and back between two locations at high
speed. This slows down file access a lot.
• If you experience a low frame rate (the mouse cursor is jumping rather
than moving smoothly), try to reduce the size of the ’Maximal Window
Width’ in the Preferences to something like 1280. On very large screens
you can set the ’Target resolution smaller than’ to 4 (Default: 2) to help.
• If VAST slows down considerably after using it for a while, check if the
RAM of your computer is full (check the RAM usage indicator in the
upper right corner of the VAST main window, see section 3.1.5). Once
memory is full, Windows might swap parts of the data to the hard drive,
which can slow down processing a lot. To fix that problem tell VAST to
use less memory by REDUCING the maximal RAM cache memory sizes
in the Preferences dialog.
A.4
Setting up VAST with a Wacom screen
When it comes to fast and accurate manual painting on a computer, tablets and
in particular tablet screens can improve productivity significantly. We are using
various Wacom Cintiq screen tablets. While the larger Cintiqs have a better
screen, they can be expensive and bulky. I personally prefer the Cintiq 13HD,
which can be laid flat onto a desk and close to the user.
An alternative could be tablet laptops, if they fulfill the system requirements
for VAST. We tried the Asus EEE Slate EP121; it works but it is a bit slow,
and pen button presses tend to be unreliable. Newer tablets like the Microsoft
Surface Pro might have improved performance, but some might not support
two-button pens.
Wacom tablets come with driver software which lets you configure the pen
buttons for each program. For optimal workflow in VAST I find it most useful
to have ’erase’ on one pen button and ’change tooltip size’ on the other. For
A.4. SETTING UP VAST WITH A WACOM SCREEN
35
this it is easiest to set one pen button to ’right click’ and the other one to
’middle click’. If that does not work for your system (sometimes Windows uses
the Right Click event in a special way, for example), you can configure the pen
buttons to simulate equivalent key presses (see Appendix A.5).
If you see brief circular animations when you use the pen and drawing is
delayed, you should go to ’Pen and Touch’ in the Windows Control Panel and
switch off ’Flicks’. On the tab ’Flicks’, uncheck ’Use flicks to perform common
actions quickly and easily’ and click ’Apply’.
36
A.5
APPENDIX A. FAQ AND TROUBLE SHOOTING
Keyboard Shortcuts in VAST
You can open a window which lists all the keyboard shortcuts in VAST from
the main menu under Window / Keyboard Shortcuts. Here is a summary.
SHIFT or ENTER
CTRL or INSERT
TAB or \|
‘~ or DELETE
H or L
Pick segment color by mouse click
Temporarily go to ’move’ mode
Move mode: Click left and move pen up/down to zoom;
Paint mode: Click left and move pen up/down to change pen diameter
Paint mode: Erase mode
Hides segmentation while held down
Table A.1: Mode Modifiers (hold down)
UP or A
DOWN or Z
PAGEUP or S
PAGEDOWN or X
One slice up
One slice down
MAXPAINTDEPTH slices up
MAXPAINTDEPTH slices down
Table A.2: Slice Navigation
MAXPAINTDEPTH is the value set under ’Max Paint Depth’ in the ’Drawing
Properties’ dialog.
-_
=+
N or Keypad /
M or Keypad *
F
I, O, P
HOME
,< / .>
Decrease tooltip diameter
Increase tooltip diameter
Zoom out
Zoom in
Flash selected segment
Set paint mode to Paint all, Background, Parent
Go to anchor point of selected segment
Select previous / next segment in recently selected list
Table A.3: Other Controls
A.6. TERMS OF USAGE AND PRIVACY STATEMENT
A.6
37
Terms of Usage and Privacy Statement
This version of VAST (’the software’) is free of charge and may be distributed
freely, but not sold. Commercial usage is allowed.
You are using this software at your own risk. Even though it has been tested
extensively, it is not free of bugs. Please keep backup copies of your data.
THE SOFTWARE IS PROVIDED ”AS IS”, WITHOUT WARRANTY OF
ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
IN THE SOFTWARE.
VAST does not collect usage statistics or other data. In particular, it does
not transmit any of your image or segmentation data anywhere.
This software uses easyzlib.c, which is based on the zlib library by Jeanloup Gailly and Mark Adler.
38
APPENDIX A. FAQ AND TROUBLE SHOOTING
Appendix B
Technical Information
B.1
Size limitations
Maximal file size for EM images and segmentations: Limited by maximal file size on disk; theoretical maximum 264 bytes.
Largest EM stack that has been imported into VAST so far: ∼ 4 Terabytes
Maximal number of labels supported: currently 216 − 1 (but needs lots of
RAM; 600 bytes per label).
Largest image supported (at full resolution): (231 − 1) · (231 − 1)
B.2
Supported file formats for importing / exporting
Importing of EM images
Currently 8 bit graylevel and 24-bit RGB image stacks are supported.
Stacks and tiled stacks: .JPG, .PNG, .TIF
3D Volume files:
.NII (NIfTI); will be converted to 8 bits when imported
Importing from .TIF images under Windows 7 can result in an image stack
that has only 16 gray levels instead of 256, because of a bug in the Windows
GDI+ TIFF routines. Consider converting the TIFF images to PNG before
importing.
Exporting of EM images
Stacks and tiled stacks:
.PNG 8-bit indexed, .TIF uncompressed, .RAW
Importing of segmentations
Segmentations can be imported from RGB .TIF and .PNG image stacks. The
segment number for each pixel is encoded in the RGB value of the image as
39
40
APPENDIX B. TECHNICAL INFORMATION
follows: Bits 0-7 of the label number are expected in the blue channel, bits 8-15
in the green channel, and bits 16-23 in the red channel. This is the same format
used for exporting segmentations to image stacks (see below). Please be aware
that VAST can currently only handle 16-bit segmentations.
Exporting of segmentations
When exporting segmentations, the available file formats depend on the range
of segment numbers used. For example, if the highest segment number is greater
than 255, 8 bit indexed file formats will not be available. In that case the label
numbers will be encoded into the color channels (for example for RGB, bits 0-7
of the label number will be put into the blue channel, bits 8-15 into the green
channel, and bits 16-23 into the red channel).
Stacks and tiled stacks: .PNG, .TIF uncompressed, .RAW
3D Volume files:
.NII (NIfTI) 8 bit only (currently unsupported)
Exporting of screen shots
Stacks and tiled stacks:
3D Volume files:
.PNG, .TIF uncompressed, .RAW (all 24 bit RGB)
.NII (Nifti) 8 bit only (currently unsupported)
Bibliography
[1] Cardona, A., Saalfeld, S., Preibisch, S., Schmid, B., Cheng,
A., Pulokas, J., Tomancak, P. and Hartenstein, V.: An Integrated Micro- and Macroarchitectural Analysis of the Drosophila Brain
by Computer-Assisted Serial Section Electron Microscopy, PLoS Biology,
8(10), (2010), 1-17 (e1000502).
[2] Cardona A, Saalfeld S, Schindelin J, Arganda-Carreras I,
Preibisch S, et al.: TrakEM2 Software for Neural Circuit Reconstruction, PLoS ONE, 7(6):e38011, (2012), doi:10.1371/journal.pone.0038011.
[3] Peng, H., Ruan, Z., Long, F., Simpson, J.H. and Myers, E.W.: V3D
enables real-time 3D visualization and quantitative analysis of large-scale
biological image data sets, Nature Biotechnology, 28(4), (2010), 348-353.
[4] Saalfeld, S., Fetter, R., Cardona, A., and Tomancak, P.: Elastic
volume reconstruction from series of ultra-thin microscopy sections, Nature
Methods, 9(7), (2012), 717-720.
41
42
BIBLIOGRAPHY